Working for a library on rare fonts I've found Tesseract fantastic for custom training. It certainly beats Abbyy from 10 years ago - maybe a low bar to clear.

Working for a library on rare fonts I've found Tesseract fantastic for custom training. It certainly beats Abbyy from 10 years ago - maybe a low bar to clear. I had to spend some time setting up labeling then did some supplemental training on UB-Mannheim datasets. Tesseract is the only OCR FOSS solution that has reasonable performance.